ºÚÁÏרÇø

From Linux to USV Swarms: Learning Maritime Autonomy with the BlueBoat Simulator

The Carnegie Mellon Robotics Academy’s branch of the  gives learners hands-on experience with the technologies used to operate autonomous maritime systems. Based on the , the project simulates an unmanned surface vehicle, or USV. These vessels can be remotely operated or programmed to navigate autonomously for hydrographic surveying, environmental monitoring, inspection, search, and other maritime applications.

Learners begin by manually undocking a single BlueBoat. They progress through autonomous navigation, geofencing, vehicle monitoring, underwater mapping, and multi-vehicle search missions.

Exploring the robotics software stack

The simulator combines several widely used robotics technologies:

• Linux provides the operating environment. Learners navigate directories, run scripts, manage processes, and work across several terminal windows.

• Docker packages the project’s software dependencies into a consistent computing environment.

• Gazebo Sim models the lake, boat, docks, buoys, waves, underwater terrain, sensors, and hazards.

• ArduPilot Software in the Loop, or SITL, runs the vehicle’s autopilot software on the computer.

• QGroundControl allows learners to create waypoint routes, survey patterns, and geofences while monitoring vehicle status.

• ROS 2 carries motor commands, odometry, position data, and simulated sensor readings between components.

• RViz visualizes the underwater mapping data collected during survey missions.

The project’s Docker environment includes ROS 2 Humble, Gazebo Garden, ArduPilot, Python libraries, graphics support, and visualization tools. Learners gain experience with containerized software while working with an authentic robotics workflow.

A mission-based learning progression

Mission 0: Software Setup and Manual Undocking: Learners start the Docker container, launch Gazebo and ArduPilot, connect QGroundControl, and configure a game controller. They then arm the BlueBoat and manually drive it away from the dock. This mission introduces the complete software stack, basic Linux commands, vehicle modes, telemetry, and geofenced areas around the dock and buoy.

Mission 1: Buoy and Back: In Mission 1a, learners create and upload an autonomous waypoint mission that travels around a buoy. They manually clear the dock, switch the BlueBoat into Auto mode, and monitor its progress. Mission 1b adds waves and current. Learners observe course drift, monitor the vehicle near a waypoint, and adjust autonomy settings to improve its ability to maintain position.

Mission 2: Channel and Narrow Channel: Mission 2a asks learners to plan a route around an island. They create exclusion zones to protect the vehicle from known navigational hazards. Mission 2b introduces a narrow channel that requires closer operator involvement. Learners plan the autonomous portions of the route and transition among Manual, Hold, and Auto modes as conditions change. These activities develop skills in mission planning, geofencing, supervised autonomy, and operator intervention.

Mission 3: Underwater Mapping: Learners create a lawnmower-style survey pattern over part of the lake. As the BlueBoat follows the route, a simulated underwater sensor produces scan data. A ROS 2 mapping node combines the scan measurements with vehicle odometry and converts them into a three-dimensional point cloud. Learners monitor the developing lakebed map in RViz and evaluate how survey design affects coverage, overlap, and mission duration.

Coordinating a fleet of USVs

Additional missions under development expand the simulation to four BlueBoats operating on the same lake. Each vehicle has its own ArduPilot instance, system identifier, motor controls, navigation data, underwater scan, and camera stream. QGroundControl can connect to all four vehicles and allow the operator to select, plan, arm, and monitor each one.

Participants will coordinate the fleet to search buoy-bounded areas for submerged vessels and other target objects. They must divide the search area, assign survey patterns, monitor vehicle progress, and interpret the data collected by the fleet.

The simulation can distribute submerged targets and distractors among the search zones using a recorded random seed. Instructors can create new layouts and reproduce a specific scenario for evaluation or troubleshooting.

The multi-vehicle missions introduce several core swarm and fleet-management concepts:

• Task and search-area allocation

• Coordinated coverage planning

• Route deconfliction

• Multi-vehicle monitoring

• Shared situational awareness

• Mission recovery and reassignment

• Combining data from multiple platforms

Preparing for real-world autonomy

Simulation gives learners time to practice mission planning, test different strategies, study system behavior, and recover from mistakes. Missions can be repeated under consistent conditions, including the same wave settings or submerged-hazard layout.

The simulator also exposes learners to the complete operational workflow surrounding an autonomous vehicle. They manage the computing environment, launch interconnected software, configure the vehicle, develop a mission, monitor its execution, interpret sensor data, and troubleshoot problems.

The learning sequence begins with Linux, Docker, and a manual departure from the dock. It grows into autonomous surveying, underwater mapping, and the coordination of a four-vehicle USV fleet.